How Much RAM Does Your Graphics Card Really Need?

[cleeve]

[citation][nom]spectrewind[/nom] Ideally, side-by-side PNG (Lossless RGB-alpha) files would have been better for this article.[/citation]

To each his own -- I'll have to disagree with you on that one.

I think the animated version does a superlative job of highlighting the differences between the detail settings. Judging from the comments, I think a good percentage of folks would agree.

I understand your concern, and I'll look into the possibility of rollover jpgs if the CMS allows it, but I have absolutely no reservations about the capability of the animated GIFs to do the job as intended in this article, regardless of the color limitations.

Like I said, we're not looking for subtle nuances in anisotropic filtering here... the purpose of the image is to show major differences in detail settings, and it did that job better than I could have asked for.

 

[Guest]

A quality article.

I am bothered, though, that you've included a few comparisons in settings where even the 2GB cards can't deliver a playable frame rate. If the card can't deliver with even excess video RAM, the comparison doesn't mean much.

I would say a minimum frame rate of 45 or so should be the limit on whether benchmark comparisons should be seen as valid and relevant to real world gaming requirements. That would provide some headroom for when things get heavy late game. If it's not playable anyway, you may as well be comparing 0.2FPS with 0.5FPS on a Radeon 9200. The comparison isn't worth much if you're simply doing whatever it takes to fill up the video RAM and ignoring the flailing GPU, which from what I can see, in most cases here, is struggling to keep up in settings that require even 512MB, let alone 1GB.

A suggestion relating to the screenshots: could you use lossless PNG rather than GIF and have the image change on mouse hover with CSS? Or perhaps use javascript to flip between PNG files on a timer? 256 colour GIF's aren't really appropriate. Another alternative might be to convert the screenshots to greyscale where 8bit GIF would be sufficient.

Thanks for the article. Please forgive if the above seems critical.

 

[avatar_raq]

Long awaited article. I -like others- was hoping to see crossfire benchmarking, where the extra GPU power is expected to better use the extra vRAM.
Personally I'm waiting for the 2GB version of 5870 because I plan to go crossfire in the future and I plan to keep it for no less than 2 years !.

 

[mapesdhs]

Hmm, another thought, what about 3-way 5850? And 5870?

Ian.

 

[Guest]

One more thing!! - Keep in mind that under any 32bit Windows, every MB of graphics card memory results in a corresponding MB less of system RAM that Windows can use. If you have 2GB of RAM and a 1GB graphics card, Windows will only be able to use 1GB of system RAM. Better make sure you don't need more than 1GB of system RAM for Windows + your game, and any other apps/processes that might be running.

Of course Windows can only see a maximum of 3.5GB of RAM regardless of how much is installed. So if you install 4GB of RAM and use a 512MB card, you theoretically shouldn't run into the RAM mapping problem at all. If you use a 1GB graphics card in this case, Windows should still see 3GB of RAM so your penalty will only be slight. I wouldn't waste money on a graphics card with more than 1GB.

What happens if you only have 2GB of system RAM and install a graphics card with 2GB of RAM? Will Windows even boot? I don't know, I haven't tried it. That could be a fun experiment though...

 

[cleeve]

[citation][nom]jaiknow[/nom]I am bothered, though, that you've included a few comparisons in settings where even the 2GB cards can't deliver a playable frame rate. If the card can't deliver with even excess video RAM, the comparison doesn't mean much. [/citation]

Like the last page of the article says though, if you need more speed a better chipset is the way to go, not more RAM.

We tested with a 4870 to demonstrate the differences, but there's lots of faster cards out there. The 4890, GTX 275, GTX 295, the new Radeon 5800 series... not to mention CrossFire and SLI options.

 

[mapesdhs]

Am I right in thinking the 5850/5870 are not supported under XP?
Something to ponder for those considering an upgrade. If you're
using XP and don't want to upgrade (more cost), then a 4890 isn't
far off a 5850 in most cases.

I had a look at GTX275/285 prices, which haven't changed yet. The
price drops on the 4870/4890 makes them far better value than
the 275/285 now.

Ian.

 

[era86]

Very very informative for those building new systems, especially with the new cards that just dropped.

 

[Major7up]

Nice to see this settled once and for all even though we kind of have known about this for some time now.

 

[Guest]

What about MMIO when running a 2GB card on a 32bit system?

WIth 4 GB of system RAM there should be only 1.2 GB addressable RAM left for Windows and applications.

This way a 2GB card will most likely degrade performance on most modern games...

 

[bazzz]

There is a discussion where the argument was made that running a 2 gig videocard on a 3 gig system would not give reliable results as it would need more ram to make use of the video memory.

http://www.scorched3d.co.uk/phpBB3/viewtopic.php?f=12&t=5849

Have there been tests done that would support this argument or can anyone explain how main ram would limit the amount of video memory ,or would not?

 

[Guest]

I'm curious what the results would be on a 32 bit OS. All that unused video ram taking address space away from the application must have some effect.

 

[dredg98]

why cant they test on a system that 3/4 of everyone has a Dual Core there is a few amount of people that have core i7 test on a dual wether it be intel or amd i like how they use the high end systems (core i7, ddr3 ram) come on not everyone has that do tests on a dual core with 4gb of ddr2 ram and a mid range 7200 rpm hdd

 

[Guest]

Given the proliferation of dual monitors these days I would have liked to see some of these tests done on this type of setup.

 

[Guest]

Another quality article. However I am very interested in seeing what affect a crossfired 512MB card Vs a crosfired 1024 MB card scenario would look like.

Is the 512MD crossfire setup still impacted by higher resolution settings i.e 2560 and 1920.

Does the crossfire setup utilise its two seperate memory banks to equate double the memory or is it still impacted in some way by its individual cards memory limit.

Should you crossfire 2 512MB's or 2 1024MB's???

P.S. Will the 4850 also scale the same as the 4870?

 

[cleeve]

[citation][nom]dredg98[/nom]why cant they test on a system that 3/4 of everyone has a Dual Core there is a few amount of people that have core i7 test on a dual wether it be intel or amd i like how they use the high end systems (core i7, ddr3 ram) come on not everyone has that do tests on a dual core with 4gb of ddr2 ram and a mid range 7200 rpm hdd[/citation]

Because if I test it on a system like that, we can't be sure the bottlenecking is from the system or the graphics cards.

When you're targeting a specific subsystem - in this case, VRAM - you have to make sure the other factors don't influence your results.

 

[dredg98]

but those factors also improve the score a whole lot. if you have a pretty good cpu (dual core) it shouldnt be a bottleneck but they should at least try it so we can get real world results

 

[cleeve]

[citation][nom]dredg98[/nom]but those factors also improve the score a whole lot. if you have a pretty good cpu (dual core) it shouldnt be a bottleneck but they should at least try it so we can get real world results[/citation]

I hear where you're coming from dredg, but the point of the article is to demonstrate memory limitations, not to compare systems.
To do that we need a proper lab test. When we're trying to isolate a single factor it doesn't make sense to muddle up the results with other possible influences, because it becomes very difficult to find anything out for sure.

The method we used leaves it pretty clear that the only factor we're seeing affecting the performance is memory usage. Using a lower-end machine and introducing other potential bottlenecks is really counterproductive when we're trying to hone in on the influence of VRAM.

In addition, regardless of the number of CPU cores - if the game is graphics card bottlenecked, then these results are just as relevant. And when you take into account that the effect of VRAM becomes more prevalent at higher resolutions, remember that this is exactly where graphics card bottlenecking occurs...

 

[mapesdhs]

Who is it that's voting everyone -ve? I haven't seen any comments
that warrant a -ve vote, just honest questions and useful answers.
I've given every one a +ve vote on pp. 5 of the comments, though
obviously I can't do that for myself.

Thanks for the clarifications, Cleeve! Could you confirm though
whether an XP system would be able to use a 5850/5870? Checking
for drivers on amd.com, XP is listed as an option in the drivers
for these new cards.

Ian.

 

[mapesdhs]

Sorry, typo there, I meant to say XP is NOT listed as an option
in the 5850/5870 driver selection panel.

Ian.

 

[N.Broekhuijsen]

Very nice. Interesting to see the GTA4 numbers to compare them to my GTS250.....

 

[scook9]

you should re run these tests with the cards in crossfirex

I have heard that this is where you can really bottleneck due to vram due to the higher GPU bandwidth (up to twice as much)

 

[bradlei]

Great article but I agree with others in wanting to see a 5870 1gb vs 2gb, 4870 maybe not enough to test the limits. 4gb os system ram in an old single core system doesn't do much but definitely an improvment over 2 with latest quad core beasts.

 

[g00ey]

It would be interesting if they recompiled some of the older games for these newer cards so that they take advantage of these new features (such as higher RAM). For example TES Oblivion is myopic in terms of draw distances and it would probably look really good if all of the trees and foliage were drawn all the way to the distant horizon.

 

[Guest]

The benchmarked 2GB card will only give you an advantage on the current games where the GPU is able to handle the game's graphics engine. As soon as a new game comes out that is more demanding it won't be playable on any of those cards at the highest settings, therefore you will be forced to lower quality & resolution and be unable to fill the available memory. So 2GB is overkill even from a longevity standpoint.